2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_types.h"
25 #include "xfs_trans.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_itable.h"
41 #include "xfs_btree.h"
42 #include "xfs_ialloc.h"
43 #include "xfs_alloc.h"
47 #include "xfs_error.h"
48 #include "xfs_quota.h"
49 #include "xfs_utils.h"
50 #include "xfs_rtalloc.h"
51 #include "xfs_trans_space.h"
52 #include "xfs_log_priv.h"
53 #include "xfs_filestream.h"
54 #include "xfs_vnodeops.h"
62 xfs_mount_t *mp = ip->i_mount;
63 struct inode *inode = VFS_I(ip);
64 int mask = iattr->ia_valid;
72 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
77 if (mp->m_flags & XFS_MOUNT_RDONLY)
78 return XFS_ERROR(EROFS);
80 if (XFS_FORCED_SHUTDOWN(mp))
81 return XFS_ERROR(EIO);
83 code = -inode_change_ok(inode, iattr);
87 olddquot1 = olddquot2 = NULL;
91 * If disk quotas is on, we make sure that the dquots do exist on disk,
92 * before we start any other transactions. Trying to do this later
93 * is messy. We don't care to take a readlock to look at the ids
94 * in inode here, because we can't hold it across the trans_reserve.
95 * If the IDs do change before we take the ilock, we're covered
96 * because the i_*dquot fields will get updated anyway.
98 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
101 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
103 qflags |= XFS_QMOPT_UQUOTA;
105 uid = ip->i_d.di_uid;
107 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
109 qflags |= XFS_QMOPT_GQUOTA;
111 gid = ip->i_d.di_gid;
115 * We take a reference when we initialize udqp and gdqp,
116 * so it is important that we never blindly double trip on
117 * the same variable. See xfs_create() for an example.
119 ASSERT(udqp == NULL);
120 ASSERT(gdqp == NULL);
121 code = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_d.di_projid,
122 qflags, &udqp, &gdqp);
128 * For the other attributes, we acquire the inode lock and
129 * first do an error checking pass.
132 lock_flags = XFS_ILOCK_EXCL;
133 if (flags & XFS_ATTR_NOLOCK)
135 if (!(mask & ATTR_SIZE)) {
136 if ((mask != (ATTR_CTIME|ATTR_ATIME|ATTR_MTIME)) ||
137 (mp->m_flags & XFS_MOUNT_WSYNC)) {
138 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
140 if ((code = xfs_trans_reserve(tp, 0,
141 XFS_ICHANGE_LOG_RES(mp), 0,
148 if (DM_EVENT_ENABLED(ip, DM_EVENT_TRUNCATE) &&
149 !(flags & XFS_ATTR_DMI)) {
150 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
151 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, ip,
152 iattr->ia_size, 0, dmflags, NULL);
159 lock_flags |= XFS_IOLOCK_EXCL;
162 xfs_ilock(ip, lock_flags);
165 * Change file ownership. Must be the owner or privileged.
167 if (mask & (ATTR_UID|ATTR_GID)) {
169 * These IDs could have changed since we last looked at them.
170 * But, we're assured that if the ownership did change
171 * while we didn't have the inode locked, inode's dquot(s)
172 * would have changed also.
174 iuid = ip->i_d.di_uid;
175 igid = ip->i_d.di_gid;
176 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
177 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
180 * Do a quota reservation only if uid/gid is actually
183 if (XFS_IS_QUOTA_RUNNING(mp) &&
184 ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
185 (XFS_IS_GQUOTA_ON(mp) && igid != gid))) {
187 code = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
188 capable(CAP_FOWNER) ?
189 XFS_QMOPT_FORCE_RES : 0);
190 if (code) /* out of quota */
196 * Truncate file. Must have write permission and not be a directory.
198 if (mask & ATTR_SIZE) {
199 /* Short circuit the truncate case for zero length files */
200 if (iattr->ia_size == 0 &&
201 ip->i_size == 0 && ip->i_d.di_nextents == 0) {
202 xfs_iunlock(ip, XFS_ILOCK_EXCL);
203 lock_flags &= ~XFS_ILOCK_EXCL;
204 if (mask & ATTR_CTIME)
205 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
210 if (S_ISDIR(ip->i_d.di_mode)) {
211 code = XFS_ERROR(EISDIR);
213 } else if (!S_ISREG(ip->i_d.di_mode)) {
214 code = XFS_ERROR(EINVAL);
219 * Make sure that the dquots are attached to the inode.
221 code = xfs_qm_dqattach_locked(ip, 0);
226 * Now we can make the changes. Before we join the inode
227 * to the transaction, if ATTR_SIZE is set then take care of
228 * the part of the truncation that must be done without the
229 * inode lock. This needs to be done before joining the inode
230 * to the transaction, because the inode cannot be unlocked
231 * once it is a part of the transaction.
233 if (iattr->ia_size > ip->i_size) {
235 * Do the first part of growing a file: zero any data
236 * in the last block that is beyond the old EOF. We
237 * need to do this before the inode is joined to the
238 * transaction to modify the i_size.
240 code = xfs_zero_eof(ip, iattr->ia_size, ip->i_size);
242 xfs_iunlock(ip, XFS_ILOCK_EXCL);
245 * We are going to log the inode size change in this
246 * transaction so any previous writes that are beyond the on
247 * disk EOF and the new EOF that have not been written out need
248 * to be written here. If we do not write the data out, we
249 * expose ourselves to the null files problem.
251 * Only flush from the on disk size to the smaller of the in
252 * memory file size or the new size as that's the range we
253 * really care about here and prevents waiting for other data
254 * not within the range we care about here.
257 ip->i_size != ip->i_d.di_size &&
258 iattr->ia_size > ip->i_d.di_size) {
259 code = xfs_flush_pages(ip,
260 ip->i_d.di_size, iattr->ia_size,
261 XFS_B_ASYNC, FI_NONE);
264 /* wait for all I/O to complete */
268 code = xfs_itruncate_data(ip, iattr->ia_size);
271 lock_flags &= ~XFS_ILOCK_EXCL;
272 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
275 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
276 if ((code = xfs_trans_reserve(tp, 0,
277 XFS_ITRUNCATE_LOG_RES(mp), 0,
278 XFS_TRANS_PERM_LOG_RES,
279 XFS_ITRUNCATE_LOG_COUNT))) {
280 xfs_trans_cancel(tp, 0);
282 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
285 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
286 xfs_ilock(ip, XFS_ILOCK_EXCL);
288 xfs_trans_ijoin(tp, ip, lock_flags);
289 xfs_trans_ihold(tp, ip);
292 * Only change the c/mtime if we are changing the size
293 * or we are explicitly asked to change it. This handles
294 * the semantic difference between truncate() and ftruncate()
295 * as implemented in the VFS.
297 if (iattr->ia_size != ip->i_size || (mask & ATTR_CTIME))
298 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
300 if (iattr->ia_size > ip->i_size) {
301 ip->i_d.di_size = iattr->ia_size;
302 ip->i_size = iattr->ia_size;
303 if (!(flags & XFS_ATTR_DMI))
304 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
305 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
306 } else if (iattr->ia_size <= ip->i_size ||
307 (iattr->ia_size == 0 && ip->i_d.di_nextents)) {
309 * signal a sync transaction unless
310 * we're truncating an already unlinked
311 * file on a wsync filesystem
313 code = xfs_itruncate_finish(&tp, ip, iattr->ia_size,
315 ((ip->i_d.di_nlink != 0 ||
316 !(mp->m_flags & XFS_MOUNT_WSYNC))
321 * Truncated "down", so we're removing references
322 * to old data here - if we now delay flushing for
323 * a long time, we expose ourselves unduly to the
324 * notorious NULL files problem. So, we mark this
325 * vnode and flush it when the file is closed, and
326 * do not wait the usual (long) time for writeout.
328 xfs_iflags_set(ip, XFS_ITRUNCATED);
331 xfs_trans_ijoin(tp, ip, lock_flags);
332 xfs_trans_ihold(tp, ip);
336 * Change file ownership. Must be the owner or privileged.
338 if (mask & (ATTR_UID|ATTR_GID)) {
340 * CAP_FSETID overrides the following restrictions:
342 * The set-user-ID and set-group-ID bits of a file will be
343 * cleared upon successful return from chown()
345 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
346 !capable(CAP_FSETID)) {
347 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
351 * Change the ownerships and register quota modifications
352 * in the transaction.
355 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
356 ASSERT(mask & ATTR_UID);
358 olddquot1 = xfs_qm_vop_chown(tp, ip,
359 &ip->i_udquot, udqp);
361 ip->i_d.di_uid = uid;
365 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
366 ASSERT(!XFS_IS_PQUOTA_ON(mp));
367 ASSERT(mask & ATTR_GID);
369 olddquot2 = xfs_qm_vop_chown(tp, ip,
370 &ip->i_gdquot, gdqp);
372 ip->i_d.di_gid = gid;
376 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
377 timeflags |= XFS_ICHGTIME_CHG;
381 * Change file access modes.
383 if (mask & ATTR_MODE) {
384 umode_t mode = iattr->ia_mode;
386 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
389 ip->i_d.di_mode &= S_IFMT;
390 ip->i_d.di_mode |= mode & ~S_IFMT;
392 inode->i_mode &= S_IFMT;
393 inode->i_mode |= mode & ~S_IFMT;
395 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
396 timeflags |= XFS_ICHGTIME_CHG;
400 * Change file access or modified times.
402 if (mask & (ATTR_ATIME|ATTR_MTIME)) {
403 if (mask & ATTR_ATIME) {
404 inode->i_atime = iattr->ia_atime;
405 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
406 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
407 ip->i_update_core = 1;
409 if (mask & ATTR_MTIME) {
410 inode->i_mtime = iattr->ia_mtime;
411 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
412 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
413 timeflags &= ~XFS_ICHGTIME_MOD;
414 timeflags |= XFS_ICHGTIME_CHG;
416 if (tp && (mask & (ATTR_MTIME_SET|ATTR_ATIME_SET)))
417 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
421 * Change file inode change time only if ATTR_CTIME set
422 * AND we have been called by a DMI function.
425 if ((flags & XFS_ATTR_DMI) && (mask & ATTR_CTIME)) {
426 inode->i_ctime = iattr->ia_ctime;
427 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
428 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
429 ip->i_update_core = 1;
430 timeflags &= ~XFS_ICHGTIME_CHG;
434 * Send out timestamp changes that need to be set to the
435 * current time. Not done when called by a DMI function.
437 if (timeflags && !(flags & XFS_ATTR_DMI))
438 xfs_ichgtime(ip, timeflags);
440 XFS_STATS_INC(xs_ig_attrchg);
443 * If this is a synchronous mount, make sure that the
444 * transaction goes to disk before returning to the user.
445 * This is slightly sub-optimal in that truncates require
446 * two sync transactions instead of one for wsync filesystems.
447 * One for the truncate and one for the timestamps since we
448 * don't want to change the timestamps unless we're sure the
449 * truncate worked. Truncates are less than 1% of the laddis
450 * mix so this probably isn't worth the trouble to optimize.
454 if (mp->m_flags & XFS_MOUNT_WSYNC)
455 xfs_trans_set_sync(tp);
457 code = xfs_trans_commit(tp, commit_flags);
460 xfs_iunlock(ip, lock_flags);
463 * Release any dquot(s) the inode had kept before chown.
465 xfs_qm_dqrele(olddquot1);
466 xfs_qm_dqrele(olddquot2);
474 if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) &&
475 !(flags & XFS_ATTR_DMI)) {
476 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, ip, DM_RIGHT_NULL,
477 NULL, DM_RIGHT_NULL, NULL, NULL,
478 0, 0, AT_DELAY_FLAG(flags));
483 commit_flags |= XFS_TRANS_ABORT;
489 xfs_trans_cancel(tp, commit_flags);
491 if (lock_flags != 0) {
492 xfs_iunlock(ip, lock_flags);
498 * The maximum pathlen is 1024 bytes. Since the minimum file system
499 * blocksize is 512 bytes, we can get a max of 2 extents back from
502 #define SYMLINK_MAPS 2
509 xfs_mount_t *mp = ip->i_mount;
510 int pathlen = ip->i_d.di_size;
511 int nmaps = SYMLINK_MAPS;
512 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
519 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0,
520 mval, &nmaps, NULL, NULL);
524 for (n = 0; n < nmaps; n++) {
525 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
526 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
528 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0);
529 error = XFS_BUF_GETERROR(bp);
531 xfs_ioerror_alert("xfs_readlink",
532 ip->i_mount, bp, XFS_BUF_ADDR(bp));
536 if (pathlen < byte_cnt)
540 memcpy(link, XFS_BUF_PTR(bp), byte_cnt);
544 link[ip->i_d.di_size] = '\0';
556 xfs_mount_t *mp = ip->i_mount;
560 xfs_itrace_entry(ip);
562 if (XFS_FORCED_SHUTDOWN(mp))
563 return XFS_ERROR(EIO);
565 xfs_ilock(ip, XFS_ILOCK_SHARED);
567 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
568 ASSERT(ip->i_d.di_size <= MAXPATHLEN);
570 pathlen = ip->i_d.di_size;
574 if (ip->i_df.if_flags & XFS_IFINLINE) {
575 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
576 link[pathlen] = '\0';
578 error = xfs_readlink_bmap(ip, link);
582 xfs_iunlock(ip, XFS_ILOCK_SHARED);
589 * This is called to sync the inode and its data out to disk. We need to hold
590 * the I/O lock while flushing the data, and the inode lock while flushing the
591 * inode. The inode lock CANNOT be held while flushing the data, so acquire
592 * after we're done with that.
600 int log_flushed = 0, changed = 1;
602 xfs_itrace_entry(ip);
604 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
605 return XFS_ERROR(EIO);
607 /* capture size updates in I/O completion before writing the inode. */
608 error = xfs_wait_on_pages(ip, 0, -1);
610 return XFS_ERROR(error);
613 * We always need to make sure that the required inode state is safe on
614 * disk. The vnode might be clean but we still might need to force the
615 * log because of committed transactions that haven't hit the disk yet.
616 * Likewise, there could be unflushed non-transactional changes to the
617 * inode core that have to go to disk and this requires us to issue
618 * a synchronous transaction to capture these changes correctly.
620 * This code relies on the assumption that if the update_* fields
621 * of the inode are clear and the inode is unpinned then it is clean
622 * and no action is required.
624 xfs_ilock(ip, XFS_ILOCK_SHARED);
626 if (!(ip->i_update_size || ip->i_update_core)) {
628 * Timestamps/size haven't changed since last inode flush or
629 * inode transaction commit. That means either nothing got
630 * written or a transaction committed which caught the updates.
631 * If the latter happened and the transaction hasn't hit the
632 * disk yet, the inode will be still be pinned. If it is,
636 xfs_iunlock(ip, XFS_ILOCK_SHARED);
638 if (xfs_ipincount(ip)) {
639 error = _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
640 XFS_LOG_FORCE | XFS_LOG_SYNC,
644 * If the inode is not pinned and nothing has changed
645 * we don't need to flush the cache.
651 * Kick off a transaction to log the inode core to get the
652 * updates. The sync transaction will also force the log.
654 xfs_iunlock(ip, XFS_ILOCK_SHARED);
655 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
656 error = xfs_trans_reserve(tp, 0,
657 XFS_FSYNC_TS_LOG_RES(ip->i_mount), 0, 0, 0);
659 xfs_trans_cancel(tp, 0);
662 xfs_ilock(ip, XFS_ILOCK_EXCL);
665 * Note - it's possible that we might have pushed ourselves out
666 * of the way during trans_reserve which would flush the inode.
667 * But there's no guarantee that the inode buffer has actually
668 * gone out yet (it's delwri). Plus the buffer could be pinned
669 * anyway if it's part of an inode in another recent
670 * transaction. So we play it safe and fire off the
671 * transaction anyway.
673 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
674 xfs_trans_ihold(tp, ip);
675 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
676 xfs_trans_set_sync(tp);
677 error = _xfs_trans_commit(tp, 0, &log_flushed);
679 xfs_iunlock(ip, XFS_ILOCK_EXCL);
682 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
684 * If the log write didn't issue an ordered tag we need
685 * to flush the disk cache for the data device now.
688 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
691 * If this inode is on the RT dev we need to flush that
694 if (XFS_IS_REALTIME_INODE(ip))
695 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
702 * This is called by xfs_inactive to free any blocks beyond eof
703 * when the link count isn't zero and by xfs_dm_punch_hole() when
704 * punching a hole to EOF.
714 xfs_fileoff_t end_fsb;
715 xfs_fileoff_t last_fsb;
716 xfs_filblks_t map_len;
718 xfs_bmbt_irec_t imap;
719 int use_iolock = (flags & XFS_FREE_EOF_LOCK);
722 * Figure out if there are any blocks beyond the end
723 * of the file. If not, then there is nothing to do.
725 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
726 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
727 map_len = last_fsb - end_fsb;
732 xfs_ilock(ip, XFS_ILOCK_SHARED);
733 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
734 NULL, 0, &imap, &nimaps, NULL, NULL);
735 xfs_iunlock(ip, XFS_ILOCK_SHARED);
737 if (!error && (nimaps != 0) &&
738 (imap.br_startblock != HOLESTARTBLOCK ||
739 ip->i_delayed_blks)) {
741 * Attach the dquots to the inode up front.
743 error = xfs_qm_dqattach(ip, 0);
748 * There are blocks after the end of file.
749 * Free them up now by truncating the file to
752 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
755 * Do the xfs_itruncate_start() call before
756 * reserving any log space because
757 * itruncate_start will call into the buffer
759 * do that within a transaction.
762 xfs_ilock(ip, XFS_IOLOCK_EXCL);
763 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
766 xfs_trans_cancel(tp, 0);
768 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
772 error = xfs_trans_reserve(tp, 0,
773 XFS_ITRUNCATE_LOG_RES(mp),
774 0, XFS_TRANS_PERM_LOG_RES,
775 XFS_ITRUNCATE_LOG_COUNT);
777 ASSERT(XFS_FORCED_SHUTDOWN(mp));
778 xfs_trans_cancel(tp, 0);
779 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
783 xfs_ilock(ip, XFS_ILOCK_EXCL);
784 xfs_trans_ijoin(tp, ip,
787 xfs_trans_ihold(tp, ip);
789 error = xfs_itruncate_finish(&tp, ip,
794 * If we get an error at this point we
795 * simply don't bother truncating the file.
799 (XFS_TRANS_RELEASE_LOG_RES |
802 error = xfs_trans_commit(tp,
803 XFS_TRANS_RELEASE_LOG_RES);
805 xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)
812 * Free a symlink that has blocks associated with it.
815 xfs_inactive_symlink_rmt(
823 xfs_fsblock_t first_block;
824 xfs_bmap_free_t free_list;
827 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
835 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
837 * We're freeing a symlink that has some
838 * blocks allocated to it. Free the
839 * blocks here. We know that we've got
840 * either 1 or 2 extents and that we can
841 * free them all in one bunmapi call.
843 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
844 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
845 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
846 ASSERT(XFS_FORCED_SHUTDOWN(mp));
847 xfs_trans_cancel(tp, 0);
852 * Lock the inode, fix the size, and join it to the transaction.
853 * Hold it so in the normal path, we still have it locked for
854 * the second transaction. In the error paths we need it
855 * held so the cancel won't rele it, see below.
857 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
858 size = (int)ip->i_d.di_size;
860 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
861 xfs_trans_ihold(tp, ip);
862 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
864 * Find the block(s) so we can inval and unmap them.
867 xfs_bmap_init(&free_list, &first_block);
868 nmaps = ARRAY_SIZE(mval);
869 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
870 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
874 * Invalidate the block(s).
876 for (i = 0; i < nmaps; i++) {
877 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
878 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
879 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
880 xfs_trans_binval(tp, bp);
883 * Unmap the dead block(s) to the free_list.
885 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
886 &first_block, &free_list, NULL, &done)))
890 * Commit the first transaction. This logs the EFI and the inode.
892 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
895 * The transaction must have been committed, since there were
896 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
897 * The new tp has the extent freeing and EFDs.
901 * The first xact was committed, so add the inode to the new one.
902 * Mark it dirty so it will be logged and moved forward in the log as
903 * part of every commit.
905 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
906 xfs_trans_ihold(tp, ip);
907 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
909 * Get a new, empty transaction to return to our caller.
911 ntp = xfs_trans_dup(tp);
913 * Commit the transaction containing extent freeing and EFDs.
914 * If we get an error on the commit here or on the reserve below,
915 * we need to unlock the inode since the new transaction doesn't
916 * have the inode attached.
918 error = xfs_trans_commit(tp, 0);
921 ASSERT(XFS_FORCED_SHUTDOWN(mp));
925 * transaction commit worked ok so we can drop the extra ticket
926 * reference that we gained in xfs_trans_dup()
928 xfs_log_ticket_put(tp->t_ticket);
931 * Remove the memory for extent descriptions (just bookkeeping).
933 if (ip->i_df.if_bytes)
934 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
935 ASSERT(ip->i_df.if_bytes == 0);
937 * Put an itruncate log reservation in the new transaction
940 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
941 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
942 ASSERT(XFS_FORCED_SHUTDOWN(mp));
946 * Return with the inode locked but not joined to the transaction.
952 xfs_bmap_cancel(&free_list);
955 * Have to come here with the inode locked and either
956 * (held and in the transaction) or (not in the transaction).
957 * If the inode isn't held then cancel would iput it, but
958 * that's wrong since this is inactive and the vnode ref
959 * count is 0 already.
960 * Cancel won't do anything to the inode if held, but it still
961 * needs to be locked until the cancel is done, if it was
962 * joined to the transaction.
964 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
965 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
972 xfs_inactive_symlink_local(
978 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
980 * We're freeing a symlink which fit into
981 * the inode. Just free the memory used
982 * to hold the old symlink.
984 error = xfs_trans_reserve(*tpp, 0,
985 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
986 0, XFS_TRANS_PERM_LOG_RES,
987 XFS_ITRUNCATE_LOG_COUNT);
990 xfs_trans_cancel(*tpp, 0);
994 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
997 * Zero length symlinks _can_ exist.
999 if (ip->i_df.if_bytes > 0) {
1000 xfs_idata_realloc(ip,
1001 -(ip->i_df.if_bytes),
1003 ASSERT(ip->i_df.if_bytes == 0);
1017 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1020 ASSERT(ip->i_d.di_forkoff != 0);
1021 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1022 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1026 error = xfs_attr_inactive(ip);
1030 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1031 error = xfs_trans_reserve(tp, 0,
1032 XFS_IFREE_LOG_RES(mp),
1033 0, XFS_TRANS_PERM_LOG_RES,
1034 XFS_INACTIVE_LOG_COUNT);
1038 xfs_ilock(ip, XFS_ILOCK_EXCL);
1039 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1040 xfs_trans_ihold(tp, ip);
1041 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1043 ASSERT(ip->i_d.di_anextents == 0);
1049 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1050 xfs_trans_cancel(tp, 0);
1053 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1061 xfs_mount_t *mp = ip->i_mount;
1064 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1067 /* If this is a read-only mount, don't do this (would generate I/O) */
1068 if (mp->m_flags & XFS_MOUNT_RDONLY)
1071 if (!XFS_FORCED_SHUTDOWN(mp)) {
1075 * If we are using filestreams, and we have an unlinked
1076 * file that we are processing the last close on, then nothing
1077 * will be able to reopen and write to this file. Purge this
1078 * inode from the filestreams cache so that it doesn't delay
1079 * teardown of the inode.
1081 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1082 xfs_filestream_deassociate(ip);
1085 * If we previously truncated this file and removed old data
1086 * in the process, we want to initiate "early" writeout on
1087 * the last close. This is an attempt to combat the notorious
1088 * NULL files problem which is particularly noticable from a
1089 * truncate down, buffered (re-)write (delalloc), followed by
1090 * a crash. What we are effectively doing here is
1091 * significantly reducing the time window where we'd otherwise
1092 * be exposed to that problem.
1094 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1095 if (truncated && VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0)
1096 xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE);
1099 if (ip->i_d.di_nlink != 0) {
1100 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1101 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
1102 ip->i_delayed_blks > 0)) &&
1103 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1104 (!(ip->i_d.di_flags &
1105 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1106 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1118 * This is called when the vnode reference count for the vnode
1119 * goes to zero. If the file has been unlinked, then it must
1120 * now be truncated. Also, we clear all of the read-ahead state
1121 * kept for the inode here since the file is now closed.
1127 xfs_bmap_free_t free_list;
1128 xfs_fsblock_t first_block;
1135 xfs_itrace_entry(ip);
1138 * If the inode is already free, then there can be nothing
1141 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
1142 ASSERT(ip->i_df.if_real_bytes == 0);
1143 ASSERT(ip->i_df.if_broot_bytes == 0);
1144 return VN_INACTIVE_CACHE;
1148 * Only do a truncate if it's a regular file with
1149 * some actual space in it. It's OK to look at the
1150 * inode's fields without the lock because we're the
1151 * only one with a reference to the inode.
1153 truncate = ((ip->i_d.di_nlink == 0) &&
1154 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1155 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
1156 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1160 if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY))
1161 XFS_SEND_DESTROY(mp, ip, DM_RIGHT_NULL);
1165 /* If this is a read-only mount, don't do this (would generate I/O) */
1166 if (mp->m_flags & XFS_MOUNT_RDONLY)
1169 if (ip->i_d.di_nlink != 0) {
1170 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1171 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
1172 ip->i_delayed_blks > 0)) &&
1173 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1174 (!(ip->i_d.di_flags &
1175 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1176 (ip->i_delayed_blks != 0)))) {
1177 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1179 return VN_INACTIVE_CACHE;
1184 ASSERT(ip->i_d.di_nlink == 0);
1186 error = xfs_qm_dqattach(ip, 0);
1188 return VN_INACTIVE_CACHE;
1190 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1193 * Do the xfs_itruncate_start() call before
1194 * reserving any log space because itruncate_start
1195 * will call into the buffer cache and we can't
1196 * do that within a transaction.
1198 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1200 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1202 xfs_trans_cancel(tp, 0);
1203 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1204 return VN_INACTIVE_CACHE;
1207 error = xfs_trans_reserve(tp, 0,
1208 XFS_ITRUNCATE_LOG_RES(mp),
1209 0, XFS_TRANS_PERM_LOG_RES,
1210 XFS_ITRUNCATE_LOG_COUNT);
1212 /* Don't call itruncate_cleanup */
1213 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1214 xfs_trans_cancel(tp, 0);
1215 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1216 return VN_INACTIVE_CACHE;
1219 xfs_ilock(ip, XFS_ILOCK_EXCL);
1220 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1221 xfs_trans_ihold(tp, ip);
1224 * normally, we have to run xfs_itruncate_finish sync.
1225 * But if filesystem is wsync and we're in the inactive
1226 * path, then we know that nlink == 0, and that the
1227 * xaction that made nlink == 0 is permanently committed
1228 * since xfs_remove runs as a synchronous transaction.
1230 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1231 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1234 xfs_trans_cancel(tp,
1235 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1236 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1237 return VN_INACTIVE_CACHE;
1239 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1242 * If we get an error while cleaning up a
1243 * symlink we bail out.
1245 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1246 xfs_inactive_symlink_rmt(ip, &tp) :
1247 xfs_inactive_symlink_local(ip, &tp);
1251 return VN_INACTIVE_CACHE;
1254 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1255 xfs_trans_ihold(tp, ip);
1257 error = xfs_trans_reserve(tp, 0,
1258 XFS_IFREE_LOG_RES(mp),
1259 0, XFS_TRANS_PERM_LOG_RES,
1260 XFS_INACTIVE_LOG_COUNT);
1262 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1263 xfs_trans_cancel(tp, 0);
1264 return VN_INACTIVE_CACHE;
1267 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1268 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1269 xfs_trans_ihold(tp, ip);
1273 * If there are attributes associated with the file
1274 * then blow them away now. The code calls a routine
1275 * that recursively deconstructs the attribute fork.
1276 * We need to just commit the current transaction
1277 * because we can't use it for xfs_attr_inactive().
1279 if (ip->i_d.di_anextents > 0) {
1280 error = xfs_inactive_attrs(ip, &tp);
1282 * If we got an error, the transaction is already
1283 * cancelled, and the inode is unlocked. Just get out.
1286 return VN_INACTIVE_CACHE;
1287 } else if (ip->i_afp) {
1288 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1294 xfs_bmap_init(&free_list, &first_block);
1295 error = xfs_ifree(tp, ip, &free_list);
1298 * If we fail to free the inode, shut down. The cancel
1299 * might do that, we need to make sure. Otherwise the
1300 * inode might be lost for a long time or forever.
1302 if (!XFS_FORCED_SHUTDOWN(mp)) {
1304 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1305 error, mp->m_fsname);
1306 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1308 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1311 * Credit the quota account(s). The inode is gone.
1313 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1316 * Just ignore errors at this point. There is nothing we can
1317 * do except to try to keep going. Make sure it's not a silent
1320 error = xfs_bmap_finish(&tp, &free_list, &committed);
1322 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1323 "xfs_bmap_finish() returned error %d", error);
1324 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1326 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1327 "xfs_trans_commit() returned error %d", error);
1331 * Release the dquots held by inode, if any.
1333 xfs_qm_dqdetach(ip);
1334 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1337 return VN_INACTIVE_CACHE;
1341 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1342 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1343 * ci_name->name will point to a the actual name (caller must free) or
1344 * will be set to NULL if an exact match is found.
1349 struct xfs_name *name,
1351 struct xfs_name *ci_name)
1357 xfs_itrace_entry(dp);
1359 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1360 return XFS_ERROR(EIO);
1362 lock_mode = xfs_ilock_map_shared(dp);
1363 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
1364 xfs_iunlock_map_shared(dp, lock_mode);
1369 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp, 0);
1373 xfs_itrace_ref(*ipp);
1378 kmem_free(ci_name->name);
1387 struct xfs_name *name,
1393 int is_dir = S_ISDIR(mode);
1394 struct xfs_mount *mp = dp->i_mount;
1395 struct xfs_inode *ip = NULL;
1396 struct xfs_trans *tp = NULL;
1398 xfs_bmap_free_t free_list;
1399 xfs_fsblock_t first_block;
1400 boolean_t unlock_dp_on_error = B_FALSE;
1404 struct xfs_dquot *udqp = NULL;
1405 struct xfs_dquot *gdqp = NULL;
1410 xfs_itrace_entry(dp);
1412 if (XFS_FORCED_SHUTDOWN(mp))
1413 return XFS_ERROR(EIO);
1415 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
1416 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1417 dp, DM_RIGHT_NULL, NULL,
1418 DM_RIGHT_NULL, name->name, NULL,
1425 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1426 prid = dp->i_d.di_projid;
1431 * Make sure that we have allocated dquot(s) on disk.
1433 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
1434 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
1440 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1441 log_res = XFS_MKDIR_LOG_RES(mp);
1442 log_count = XFS_MKDIR_LOG_COUNT;
1443 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1445 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1446 log_res = XFS_CREATE_LOG_RES(mp);
1447 log_count = XFS_CREATE_LOG_COUNT;
1448 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1451 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1454 * Initially assume that the file does not exist and
1455 * reserve the resources for that case. If that is not
1456 * the case we'll drop the one we have and get a more
1457 * appropriate transaction later.
1459 error = xfs_trans_reserve(tp, resblks, log_res, 0,
1460 XFS_TRANS_PERM_LOG_RES, log_count);
1461 if (error == ENOSPC) {
1462 /* flush outstanding delalloc blocks and retry */
1463 xfs_flush_inodes(dp);
1464 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1465 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1467 if (error == ENOSPC) {
1468 /* No space at all so try a "no-allocation" reservation */
1470 error = xfs_trans_reserve(tp, 0, log_res, 0,
1471 XFS_TRANS_PERM_LOG_RES, log_count);
1475 goto out_trans_cancel;
1478 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1479 unlock_dp_on_error = B_TRUE;
1482 * Check for directory link count overflow.
1484 if (is_dir && dp->i_d.di_nlink >= XFS_MAXLINK) {
1485 error = XFS_ERROR(EMLINK);
1486 goto out_trans_cancel;
1489 xfs_bmap_init(&free_list, &first_block);
1492 * Reserve disk quota and the inode.
1494 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
1496 goto out_trans_cancel;
1498 error = xfs_dir_canenter(tp, dp, name, resblks);
1500 goto out_trans_cancel;
1503 * A newly created regular or special file just has one directory
1504 * entry pointing to them, but a directory also the "." entry
1505 * pointing to itself.
1507 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, credp,
1508 prid, resblks > 0, &ip, &committed);
1510 if (error == ENOSPC)
1511 goto out_trans_cancel;
1512 goto out_trans_abort;
1516 * At this point, we've gotten a newly allocated inode.
1517 * It is locked (and joined to the transaction).
1520 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1523 * Now we join the directory inode to the transaction. We do not do it
1524 * earlier because xfs_dir_ialloc might commit the previous transaction
1525 * (and release all the locks). An error from here on will result in
1526 * the transaction cancel unlocking dp so don't do it explicitly in the
1530 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1531 unlock_dp_on_error = B_FALSE;
1533 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1534 &first_block, &free_list, resblks ?
1535 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1537 ASSERT(error != ENOSPC);
1538 goto out_trans_abort;
1540 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1541 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1544 error = xfs_dir_init(tp, ip, dp);
1546 goto out_bmap_cancel;
1548 error = xfs_bumplink(tp, dp);
1550 goto out_bmap_cancel;
1554 * If this is a synchronous mount, make sure that the
1555 * create transaction goes to disk before returning to
1558 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1559 xfs_trans_set_sync(tp);
1562 * Attach the dquot(s) to the inodes and modify them incore.
1563 * These ids of the inode couldn't have changed since the new
1564 * inode has been locked ever since it was created.
1566 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
1569 * xfs_trans_commit normally decrements the vnode ref count
1570 * when it unlocks the inode. Since we want to return the
1571 * vnode to the caller, we bump the vnode ref count now.
1575 error = xfs_bmap_finish(&tp, &free_list, &committed);
1577 goto out_abort_rele;
1579 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1585 xfs_qm_dqrele(udqp);
1586 xfs_qm_dqrele(gdqp);
1590 /* Fallthrough to std_return with error = 0 */
1592 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) {
1593 XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, dp, DM_RIGHT_NULL,
1594 ip, DM_RIGHT_NULL, name->name, NULL, mode,
1601 xfs_bmap_cancel(&free_list);
1603 cancel_flags |= XFS_TRANS_ABORT;
1605 xfs_trans_cancel(tp, cancel_flags);
1607 xfs_qm_dqrele(udqp);
1608 xfs_qm_dqrele(gdqp);
1610 if (unlock_dp_on_error)
1611 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1617 * Wait until after the current transaction is aborted to
1618 * release the inode. This prevents recursive transactions
1619 * and deadlocks from xfs_inactive.
1621 xfs_bmap_cancel(&free_list);
1622 cancel_flags |= XFS_TRANS_ABORT;
1623 xfs_trans_cancel(tp, cancel_flags);
1625 unlock_dp_on_error = B_FALSE;
1631 int xfs_small_retries;
1632 int xfs_middle_retries;
1633 int xfs_lots_retries;
1634 int xfs_lock_delays;
1638 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1642 xfs_lock_inumorder(int lock_mode, int subclass)
1644 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1645 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
1646 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
1647 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
1653 * The following routine will lock n inodes in exclusive mode.
1654 * We assume the caller calls us with the inodes in i_ino order.
1656 * We need to detect deadlock where an inode that we lock
1657 * is in the AIL and we start waiting for another inode that is locked
1658 * by a thread in a long running transaction (such as truncate). This can
1659 * result in deadlock since the long running trans might need to wait
1660 * for the inode we just locked in order to push the tail and free space
1669 int attempts = 0, i, j, try_lock;
1672 ASSERT(ips && (inodes >= 2)); /* we need at least two */
1678 for (; i < inodes; i++) {
1681 if (i && (ips[i] == ips[i-1])) /* Already locked */
1685 * If try_lock is not set yet, make sure all locked inodes
1686 * are not in the AIL.
1687 * If any are, set try_lock to be used later.
1691 for (j = (i - 1); j >= 0 && !try_lock; j--) {
1692 lp = (xfs_log_item_t *)ips[j]->i_itemp;
1693 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1700 * If any of the previous locks we have locked is in the AIL,
1701 * we must TRY to get the second and subsequent locks. If
1702 * we can't get any, we must release all we have
1707 /* try_lock must be 0 if i is 0. */
1709 * try_lock means we have an inode locked
1710 * that is in the AIL.
1713 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
1717 * Unlock all previous guys and try again.
1718 * xfs_iunlock will try to push the tail
1719 * if the inode is in the AIL.
1722 for(j = i - 1; j >= 0; j--) {
1725 * Check to see if we've already
1726 * unlocked this one.
1727 * Not the first one going back,
1728 * and the inode ptr is the same.
1730 if ((j != (i - 1)) && ips[j] ==
1734 xfs_iunlock(ips[j], lock_mode);
1737 if ((attempts % 5) == 0) {
1738 delay(1); /* Don't just spin the CPU */
1748 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
1754 if (attempts < 5) xfs_small_retries++;
1755 else if (attempts < 100) xfs_middle_retries++;
1756 else xfs_lots_retries++;
1764 * xfs_lock_two_inodes() can only be used to lock one type of lock
1765 * at a time - the iolock or the ilock, but not both at once. If
1766 * we lock both at once, lockdep will report false positives saying
1767 * we have violated locking orders.
1770 xfs_lock_two_inodes(
1779 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1780 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
1781 ASSERT(ip0->i_ino != ip1->i_ino);
1783 if (ip0->i_ino > ip1->i_ino) {
1790 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
1793 * If the first lock we have locked is in the AIL, we must TRY to get
1794 * the second lock. If we can't get it, we must release the first one
1797 lp = (xfs_log_item_t *)ip0->i_itemp;
1798 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1799 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
1800 xfs_iunlock(ip0, lock_mode);
1801 if ((++attempts % 5) == 0)
1802 delay(1); /* Don't just spin the CPU */
1806 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
1813 struct xfs_name *name,
1816 xfs_mount_t *mp = dp->i_mount;
1817 xfs_trans_t *tp = NULL;
1818 int is_dir = S_ISDIR(ip->i_d.di_mode);
1820 xfs_bmap_free_t free_list;
1821 xfs_fsblock_t first_block;
1828 xfs_itrace_entry(dp);
1829 xfs_itrace_entry(ip);
1831 if (XFS_FORCED_SHUTDOWN(mp))
1832 return XFS_ERROR(EIO);
1834 if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) {
1835 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dp, DM_RIGHT_NULL,
1836 NULL, DM_RIGHT_NULL, name->name, NULL,
1837 ip->i_d.di_mode, 0, 0);
1842 error = xfs_qm_dqattach(dp, 0);
1846 error = xfs_qm_dqattach(ip, 0);
1851 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
1852 log_count = XFS_DEFAULT_LOG_COUNT;
1854 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
1855 log_count = XFS_REMOVE_LOG_COUNT;
1857 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1860 * We try to get the real space reservation first,
1861 * allowing for directory btree deletion(s) implying
1862 * possible bmap insert(s). If we can't get the space
1863 * reservation then we use 0 instead, and avoid the bmap
1864 * btree insert(s) in the directory code by, if the bmap
1865 * insert tries to happen, instead trimming the LAST
1866 * block from the directory.
1868 resblks = XFS_REMOVE_SPACE_RES(mp);
1869 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
1870 XFS_TRANS_PERM_LOG_RES, log_count);
1871 if (error == ENOSPC) {
1873 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
1874 XFS_TRANS_PERM_LOG_RES, log_count);
1877 ASSERT(error != ENOSPC);
1879 goto out_trans_cancel;
1882 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
1885 * At this point, we've gotten both the directory and the entry
1889 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1892 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1895 * If we're removing a directory perform some additional validation.
1898 ASSERT(ip->i_d.di_nlink >= 2);
1899 if (ip->i_d.di_nlink != 2) {
1900 error = XFS_ERROR(ENOTEMPTY);
1901 goto out_trans_cancel;
1903 if (!xfs_dir_isempty(ip)) {
1904 error = XFS_ERROR(ENOTEMPTY);
1905 goto out_trans_cancel;
1909 xfs_bmap_init(&free_list, &first_block);
1910 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
1911 &first_block, &free_list, resblks);
1913 ASSERT(error != ENOENT);
1914 goto out_bmap_cancel;
1916 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1920 * Drop the link from ip's "..".
1922 error = xfs_droplink(tp, dp);
1924 goto out_bmap_cancel;
1927 * Drop the "." link from ip to self.
1929 error = xfs_droplink(tp, ip);
1931 goto out_bmap_cancel;
1934 * When removing a non-directory we need to log the parent
1935 * inode here. For a directory this is done implicitly
1936 * by the xfs_droplink call for the ".." entry.
1938 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1942 * Drop the link from dp to ip.
1944 error = xfs_droplink(tp, ip);
1946 goto out_bmap_cancel;
1949 * Determine if this is the last link while
1950 * we are in the transaction.
1952 link_zero = (ip->i_d.di_nlink == 0);
1955 * If this is a synchronous mount, make sure that the
1956 * remove transaction goes to disk before returning to
1959 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1960 xfs_trans_set_sync(tp);
1962 error = xfs_bmap_finish(&tp, &free_list, &committed);
1964 goto out_bmap_cancel;
1966 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1971 * If we are using filestreams, kill the stream association.
1972 * If the file is still open it may get a new one but that
1973 * will get killed on last close in xfs_close() so we don't
1974 * have to worry about that.
1976 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
1977 xfs_filestream_deassociate(ip);
1979 xfs_itrace_exit(ip);
1980 xfs_itrace_exit(dp);
1983 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) {
1984 XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, dp, DM_RIGHT_NULL,
1985 NULL, DM_RIGHT_NULL, name->name, NULL,
1986 ip->i_d.di_mode, error, 0);
1992 xfs_bmap_cancel(&free_list);
1993 cancel_flags |= XFS_TRANS_ABORT;
1995 xfs_trans_cancel(tp, cancel_flags);
2003 struct xfs_name *target_name)
2005 xfs_mount_t *mp = tdp->i_mount;
2008 xfs_bmap_free_t free_list;
2009 xfs_fsblock_t first_block;
2014 xfs_itrace_entry(tdp);
2015 xfs_itrace_entry(sip);
2017 ASSERT(!S_ISDIR(sip->i_d.di_mode));
2019 if (XFS_FORCED_SHUTDOWN(mp))
2020 return XFS_ERROR(EIO);
2022 if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) {
2023 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2026 target_name->name, NULL, 0, 0, 0);
2031 /* Return through std_return after this point. */
2033 error = xfs_qm_dqattach(sip, 0);
2037 error = xfs_qm_dqattach(tdp, 0);
2041 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2042 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2043 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
2044 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2045 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2046 if (error == ENOSPC) {
2048 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2049 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2056 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
2059 * Increment vnode ref counts since xfs_trans_commit &
2060 * xfs_trans_cancel will both unlock the inodes and
2061 * decrement the associated ref counts.
2065 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2066 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2069 * If the source has too many links, we can't make any more to it.
2071 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2072 error = XFS_ERROR(EMLINK);
2077 * If we are using project inheritance, we only allow hard link
2078 * creation in our tree when the project IDs are the same; else
2079 * the tree quota mechanism could be circumvented.
2081 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2082 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2083 error = XFS_ERROR(EXDEV);
2087 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
2091 xfs_bmap_init(&free_list, &first_block);
2093 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
2094 &first_block, &free_list, resblks);
2097 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2098 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2100 error = xfs_bumplink(tp, sip);
2105 * If this is a synchronous mount, make sure that the
2106 * link transaction goes to disk before returning to
2109 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2110 xfs_trans_set_sync(tp);
2113 error = xfs_bmap_finish (&tp, &free_list, &committed);
2115 xfs_bmap_cancel(&free_list);
2119 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2123 /* Fall through to std_return with error = 0. */
2125 if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) {
2126 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2129 target_name->name, NULL, 0, error, 0);
2134 cancel_flags |= XFS_TRANS_ABORT;
2138 xfs_trans_cancel(tp, cancel_flags);
2145 struct xfs_name *link_name,
2146 const char *target_path,
2151 xfs_mount_t *mp = dp->i_mount;
2156 xfs_bmap_free_t free_list;
2157 xfs_fsblock_t first_block;
2158 boolean_t unlock_dp_on_error = B_FALSE;
2161 xfs_fileoff_t first_fsb;
2162 xfs_filblks_t fs_blocks;
2164 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
2166 const char *cur_chunk;
2171 struct xfs_dquot *udqp, *gdqp;
2179 xfs_itrace_entry(dp);
2181 if (XFS_FORCED_SHUTDOWN(mp))
2182 return XFS_ERROR(EIO);
2185 * Check component lengths of the target path name.
2187 pathlen = strlen(target_path);
2188 if (pathlen >= MAXPATHLEN) /* total string too long */
2189 return XFS_ERROR(ENAMETOOLONG);
2191 if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) {
2192 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dp,
2193 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2194 link_name->name, target_path, 0, 0, 0);
2199 /* Return through std_return after this point. */
2202 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2203 prid = dp->i_d.di_projid;
2205 prid = (xfs_prid_t)dfltprid;
2208 * Make sure that we have allocated dquot(s) on disk.
2210 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
2211 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2215 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
2216 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2218 * The symlink will fit into the inode data fork?
2219 * There can't be any attributes so we get the whole variable part.
2221 if (pathlen <= XFS_LITINO(mp))
2224 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
2225 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
2226 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
2227 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2228 if (error == ENOSPC && fs_blocks == 0) {
2230 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
2231 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2238 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2239 unlock_dp_on_error = B_TRUE;
2242 * Check whether the directory allows new symlinks or not.
2244 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
2245 error = XFS_ERROR(EPERM);
2250 * Reserve disk quota : blocks and inode.
2252 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
2257 * Check for ability to enter directory entry, if no space reserved.
2259 error = xfs_dir_canenter(tp, dp, link_name, resblks);
2263 * Initialize the bmap freelist prior to calling either
2264 * bmapi or the directory create code.
2266 xfs_bmap_init(&free_list, &first_block);
2269 * Allocate an inode for the symlink.
2271 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT),
2272 1, 0, credp, prid, resblks > 0, &ip, NULL);
2274 if (error == ENOSPC)
2281 * An error after we've joined dp to the transaction will result in the
2282 * transaction cancel unlocking dp so don't do it explicitly in the
2286 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2287 unlock_dp_on_error = B_FALSE;
2290 * Also attach the dquot(s) to it, if applicable.
2292 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
2295 resblks -= XFS_IALLOC_SPACE_RES(mp);
2297 * If the symlink will fit into the inode, write it inline.
2299 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
2300 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
2301 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
2302 ip->i_d.di_size = pathlen;
2305 * The inode was initially created in extent format.
2307 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
2308 ip->i_df.if_flags |= XFS_IFINLINE;
2310 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
2311 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
2315 nmaps = SYMLINK_MAPS;
2317 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
2318 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
2319 &first_block, resblks, mval, &nmaps,
2326 resblks -= fs_blocks;
2327 ip->i_d.di_size = pathlen;
2328 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2330 cur_chunk = target_path;
2331 for (n = 0; n < nmaps; n++) {
2332 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
2333 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
2334 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
2335 BTOBB(byte_cnt), 0);
2336 ASSERT(bp && !XFS_BUF_GETERROR(bp));
2337 if (pathlen < byte_cnt) {
2340 pathlen -= byte_cnt;
2342 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
2343 cur_chunk += byte_cnt;
2345 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
2350 * Create the directory entry for the symlink.
2352 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
2353 &first_block, &free_list, resblks);
2356 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2357 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2360 * If this is a synchronous mount, make sure that the
2361 * symlink transaction goes to disk before returning to
2364 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2365 xfs_trans_set_sync(tp);
2369 * xfs_trans_commit normally decrements the vnode ref count
2370 * when it unlocks the inode. Since we want to return the
2371 * vnode to the caller, we bump the vnode ref count now.
2375 error = xfs_bmap_finish(&tp, &free_list, &committed);
2379 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2380 xfs_qm_dqrele(udqp);
2381 xfs_qm_dqrele(gdqp);
2383 /* Fall through to std_return with error = 0 or errno from
2384 * xfs_trans_commit */
2386 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) {
2387 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
2390 DM_RIGHT_NULL, link_name->name,
2391 target_path, 0, error, 0);
2401 xfs_bmap_cancel(&free_list);
2402 cancel_flags |= XFS_TRANS_ABORT;
2404 xfs_trans_cancel(tp, cancel_flags);
2405 xfs_qm_dqrele(udqp);
2406 xfs_qm_dqrele(gdqp);
2408 if (unlock_dp_on_error)
2409 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2420 xfs_mount_t *mp = ip->i_mount;
2424 if (!capable(CAP_SYS_ADMIN))
2425 return XFS_ERROR(EPERM);
2427 if (XFS_FORCED_SHUTDOWN(mp))
2428 return XFS_ERROR(EIO);
2430 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
2431 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
2433 xfs_trans_cancel(tp, 0);
2436 xfs_ilock(ip, XFS_ILOCK_EXCL);
2437 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2439 ip->i_d.di_dmevmask = evmask;
2440 ip->i_d.di_dmstate = state;
2442 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2444 error = xfs_trans_commit(tp, 0);
2454 xfs_itrace_entry(ip);
2456 ASSERT(!VN_MAPPED(VFS_I(ip)));
2458 /* bad inode, get out here ASAP */
2459 if (is_bad_inode(VFS_I(ip))) {
2466 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
2469 * Make sure the atime in the XFS inode is correct before freeing the
2472 xfs_synchronize_atime(ip);
2475 * If we have nothing to flush with this inode then complete the
2476 * teardown now, otherwise break the link between the xfs inode and the
2477 * linux inode and clean up the xfs inode later. This avoids flushing
2478 * the inode to disk during the delete operation itself.
2480 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
2481 * first to ensure that xfs_iunpin() will never see an xfs inode
2482 * that has a linux inode being reclaimed. Synchronisation is provided
2483 * by the i_flags_lock.
2485 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
2486 xfs_ilock(ip, XFS_ILOCK_EXCL);
2488 xfs_iflags_set(ip, XFS_IRECLAIMABLE);
2489 return xfs_reclaim_inode(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
2491 xfs_inode_set_reclaim_tag(ip);
2496 * xfs_alloc_file_space()
2497 * This routine allocates disk space for the given file.
2499 * If alloc_type == 0, this request is for an ALLOCSP type
2500 * request which will change the file size. In this case, no
2501 * DMAPI event will be generated by the call. A TRUNCATE event
2502 * will be generated later by xfs_setattr.
2504 * If alloc_type != 0, this request is for a RESVSP type
2505 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2506 * lower block boundary byte address is less than the file's
2515 xfs_alloc_file_space(
2522 xfs_mount_t *mp = ip->i_mount;
2524 xfs_filblks_t allocated_fsb;
2525 xfs_filblks_t allocatesize_fsb;
2526 xfs_extlen_t extsz, temp;
2527 xfs_fileoff_t startoffset_fsb;
2528 xfs_fsblock_t firstfsb;
2534 xfs_bmbt_irec_t imaps[1], *imapp;
2535 xfs_bmap_free_t free_list;
2536 uint qblocks, resblks, resrtextents;
2540 xfs_itrace_entry(ip);
2542 if (XFS_FORCED_SHUTDOWN(mp))
2543 return XFS_ERROR(EIO);
2545 error = xfs_qm_dqattach(ip, 0);
2550 return XFS_ERROR(EINVAL);
2552 rt = XFS_IS_REALTIME_INODE(ip);
2553 extsz = xfs_get_extsz_hint(ip);
2558 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
2559 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
2560 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
2562 /* Generate a DMAPI event if needed. */
2563 if (alloc_type != 0 && offset < ip->i_size &&
2564 (attr_flags & XFS_ATTR_DMI) == 0 &&
2565 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
2566 xfs_off_t end_dmi_offset;
2568 end_dmi_offset = offset+len;
2569 if (end_dmi_offset > ip->i_size)
2570 end_dmi_offset = ip->i_size;
2571 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, offset,
2572 end_dmi_offset - offset, 0, NULL);
2578 * Allocate file space until done or until there is an error
2581 while (allocatesize_fsb && !error) {
2585 * Determine space reservations for data/realtime.
2587 if (unlikely(extsz)) {
2588 s = startoffset_fsb;
2591 e = startoffset_fsb + allocatesize_fsb;
2592 if ((temp = do_mod(startoffset_fsb, extsz)))
2594 if ((temp = do_mod(e, extsz)))
2598 e = allocatesize_fsb;
2602 resrtextents = qblocks = (uint)(e - s);
2603 resrtextents /= mp->m_sb.sb_rextsize;
2604 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2605 quota_flag = XFS_QMOPT_RES_RTBLKS;
2608 resblks = qblocks = \
2609 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
2610 quota_flag = XFS_QMOPT_RES_REGBLKS;
2614 * Allocate and setup the transaction.
2616 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
2617 error = xfs_trans_reserve(tp, resblks,
2618 XFS_WRITE_LOG_RES(mp), resrtextents,
2619 XFS_TRANS_PERM_LOG_RES,
2620 XFS_WRITE_LOG_COUNT);
2622 * Check for running out of space
2626 * Free the transaction structure.
2628 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2629 xfs_trans_cancel(tp, 0);
2632 xfs_ilock(ip, XFS_ILOCK_EXCL);
2633 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
2638 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2639 xfs_trans_ihold(tp, ip);
2642 * Issue the xfs_bmapi() call to allocate the blocks
2644 xfs_bmap_init(&free_list, &firstfsb);
2645 error = xfs_bmapi(tp, ip, startoffset_fsb,
2646 allocatesize_fsb, bmapi_flag,
2647 &firstfsb, 0, imapp, &nimaps,
2654 * Complete the transaction
2656 error = xfs_bmap_finish(&tp, &free_list, &committed);
2661 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2662 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2667 allocated_fsb = imapp->br_blockcount;
2670 error = XFS_ERROR(ENOSPC);
2674 startoffset_fsb += allocated_fsb;
2675 allocatesize_fsb -= allocated_fsb;
2678 if (error == ENOSPC && (attr_flags & XFS_ATTR_DMI) == 0 &&
2679 DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) {
2680 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
2683 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
2685 goto retry; /* Maybe DMAPI app. has made space */
2686 /* else fall through with error from XFS_SEND_DATA */
2691 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
2692 xfs_bmap_cancel(&free_list);
2693 xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
2695 error1: /* Just cancel transaction */
2696 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2697 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2698 goto dmapi_enospc_check;
2702 * Zero file bytes between startoff and endoff inclusive.
2703 * The iolock is held exclusive and no blocks are buffered.
2705 * This function is used by xfs_free_file_space() to zero
2706 * partial blocks when the range to free is not block aligned.
2707 * When unreserving space with boundaries that are not block
2708 * aligned we round up the start and round down the end
2709 * boundaries and then use this function to zero the parts of
2710 * the blocks that got dropped during the rounding.
2713 xfs_zero_remaining_bytes(
2718 xfs_bmbt_irec_t imap;
2719 xfs_fileoff_t offset_fsb;
2720 xfs_off_t lastoffset;
2723 xfs_mount_t *mp = ip->i_mount;
2728 * Avoid doing I/O beyond eof - it's not necessary
2729 * since nothing can read beyond eof. The space will
2730 * be zeroed when the file is extended anyway.
2732 if (startoff >= ip->i_size)
2735 if (endoff > ip->i_size)
2736 endoff = ip->i_size;
2738 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
2739 XFS_IS_REALTIME_INODE(ip) ?
2740 mp->m_rtdev_targp : mp->m_ddev_targp);
2742 return XFS_ERROR(ENOMEM);
2744 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
2745 offset_fsb = XFS_B_TO_FSBT(mp, offset);
2747 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0,
2748 NULL, 0, &imap, &nimap, NULL, NULL);
2749 if (error || nimap < 1)
2751 ASSERT(imap.br_blockcount >= 1);
2752 ASSERT(imap.br_startoff == offset_fsb);
2753 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
2754 if (lastoffset > endoff)
2755 lastoffset = endoff;
2756 if (imap.br_startblock == HOLESTARTBLOCK)
2758 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2759 if (imap.br_state == XFS_EXT_UNWRITTEN)
2762 XFS_BUF_UNWRITE(bp);
2764 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
2766 error = xfs_iowait(bp);
2768 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
2769 mp, bp, XFS_BUF_ADDR(bp));
2772 memset(XFS_BUF_PTR(bp) +
2773 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
2774 0, lastoffset - offset + 1);
2779 error = xfs_iowait(bp);
2781 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
2782 mp, bp, XFS_BUF_ADDR(bp));
2791 * xfs_free_file_space()
2792 * This routine frees disk space for the given file.
2794 * This routine is only called by xfs_change_file_space
2795 * for an UNRESVSP type call.
2803 xfs_free_file_space(
2811 xfs_off_t end_dmi_offset;
2812 xfs_fileoff_t endoffset_fsb;
2814 xfs_fsblock_t firstfsb;
2815 xfs_bmap_free_t free_list;
2816 xfs_bmbt_irec_t imap;
2824 xfs_fileoff_t startoffset_fsb;
2826 int need_iolock = 1;
2830 xfs_itrace_entry(ip);
2832 error = xfs_qm_dqattach(ip, 0);
2837 if (len <= 0) /* if nothing being freed */
2839 rt = XFS_IS_REALTIME_INODE(ip);
2840 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
2841 end_dmi_offset = offset + len;
2842 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
2844 if (offset < ip->i_size && (attr_flags & XFS_ATTR_DMI) == 0 &&
2845 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
2846 if (end_dmi_offset > ip->i_size)
2847 end_dmi_offset = ip->i_size;
2848 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip,
2849 offset, end_dmi_offset - offset,
2850 AT_DELAY_FLAG(attr_flags), NULL);
2855 if (attr_flags & XFS_ATTR_NOLOCK)
2858 xfs_ilock(ip, XFS_IOLOCK_EXCL);
2859 /* wait for the completion of any pending DIOs */
2863 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
2864 ioffset = offset & ~(rounding - 1);
2866 if (VN_CACHED(VFS_I(ip)) != 0) {
2867 xfs_inval_cached_trace(ip, ioffset, -1, ioffset, -1);
2868 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
2870 goto out_unlock_iolock;
2874 * Need to zero the stuff we're not freeing, on disk.
2875 * If it's a realtime file & can't use unwritten extents then we
2876 * actually need to zero the extent edges. Otherwise xfs_bunmapi
2877 * will take care of it for us.
2879 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
2881 error = xfs_bmapi(NULL, ip, startoffset_fsb,
2882 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
2884 goto out_unlock_iolock;
2885 ASSERT(nimap == 0 || nimap == 1);
2886 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2889 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2890 block = imap.br_startblock;
2891 mod = do_div(block, mp->m_sb.sb_rextsize);
2893 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
2896 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1,
2897 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
2899 goto out_unlock_iolock;
2900 ASSERT(nimap == 0 || nimap == 1);
2901 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2902 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2904 if (mod && (mod != mp->m_sb.sb_rextsize))
2905 endoffset_fsb -= mod;
2908 if ((done = (endoffset_fsb <= startoffset_fsb)))
2910 * One contiguous piece to clear
2912 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
2915 * Some full blocks, possibly two pieces to clear
2917 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
2918 error = xfs_zero_remaining_bytes(ip, offset,
2919 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
2921 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
2922 error = xfs_zero_remaining_bytes(ip,
2923 XFS_FSB_TO_B(mp, endoffset_fsb),
2928 * free file space until done or until there is an error
2930 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2931 while (!error && !done) {
2934 * allocate and setup the transaction. Allow this
2935 * transaction to dip into the reserve blocks to ensure
2936 * the freeing of the space succeeds at ENOSPC.
2938 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
2939 tp->t_flags |= XFS_TRANS_RESERVE;
2940 error = xfs_trans_reserve(tp,
2942 XFS_WRITE_LOG_RES(mp),
2944 XFS_TRANS_PERM_LOG_RES,
2945 XFS_WRITE_LOG_COUNT);
2948 * check for running out of space
2952 * Free the transaction structure.
2954 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2955 xfs_trans_cancel(tp, 0);
2958 xfs_ilock(ip, XFS_ILOCK_EXCL);
2959 error = xfs_trans_reserve_quota(tp, mp,
2960 ip->i_udquot, ip->i_gdquot,
2961 resblks, 0, XFS_QMOPT_RES_REGBLKS);
2965 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2966 xfs_trans_ihold(tp, ip);
2969 * issue the bunmapi() call to free the blocks
2971 xfs_bmap_init(&free_list, &firstfsb);
2972 error = xfs_bunmapi(tp, ip, startoffset_fsb,
2973 endoffset_fsb - startoffset_fsb,
2974 0, 2, &firstfsb, &free_list, NULL, &done);
2980 * complete the transaction
2982 error = xfs_bmap_finish(&tp, &free_list, &committed);
2987 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2988 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2993 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
2997 xfs_bmap_cancel(&free_list);
2999 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
3000 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
3006 * xfs_change_file_space()
3007 * This routine allocates or frees disk space for the given file.
3008 * The user specified parameters are checked for alignment and size
3017 xfs_change_file_space(
3024 xfs_mount_t *mp = ip->i_mount;
3029 xfs_off_t startoffset;
3034 xfs_itrace_entry(ip);
3036 if (!S_ISREG(ip->i_d.di_mode))
3037 return XFS_ERROR(EINVAL);
3039 switch (bf->l_whence) {
3040 case 0: /*SEEK_SET*/
3042 case 1: /*SEEK_CUR*/
3043 bf->l_start += offset;
3045 case 2: /*SEEK_END*/
3046 bf->l_start += ip->i_size;
3049 return XFS_ERROR(EINVAL);
3052 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
3054 if ( (bf->l_start < 0)
3055 || (bf->l_start > XFS_MAXIOFFSET(mp))
3056 || (bf->l_start + llen < 0)
3057 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
3058 return XFS_ERROR(EINVAL);
3062 startoffset = bf->l_start;
3066 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
3068 * These calls do NOT zero the data space allocated to the file,
3069 * nor do they change the file size.
3071 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
3073 * These calls cause the new file data to be zeroed and the file
3074 * size to be changed.
3076 setprealloc = clrprealloc = 0;
3079 case XFS_IOC_RESVSP:
3080 case XFS_IOC_RESVSP64:
3081 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
3088 case XFS_IOC_UNRESVSP:
3089 case XFS_IOC_UNRESVSP64:
3090 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
3095 case XFS_IOC_ALLOCSP:
3096 case XFS_IOC_ALLOCSP64:
3097 case XFS_IOC_FREESP:
3098 case XFS_IOC_FREESP64:
3099 if (startoffset > fsize) {
3100 error = xfs_alloc_file_space(ip, fsize,
3101 startoffset - fsize, 0, attr_flags);
3106 iattr.ia_valid = ATTR_SIZE;
3107 iattr.ia_size = startoffset;
3109 error = xfs_setattr(ip, &iattr, attr_flags);
3119 return XFS_ERROR(EINVAL);
3123 * update the inode timestamp, mode, and prealloc flag bits
3125 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
3127 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
3130 xfs_trans_cancel(tp, 0);
3134 xfs_ilock(ip, XFS_ILOCK_EXCL);
3136 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3137 xfs_trans_ihold(tp, ip);
3139 if ((attr_flags & XFS_ATTR_DMI) == 0) {
3140 ip->i_d.di_mode &= ~S_ISUID;
3143 * Note that we don't have to worry about mandatory
3144 * file locking being disabled here because we only
3145 * clear the S_ISGID bit if the Group execute bit is
3146 * on, but if it was on then mandatory locking wouldn't
3147 * have been enabled.
3149 if (ip->i_d.di_mode & S_IXGRP)
3150 ip->i_d.di_mode &= ~S_ISGID;
3152 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3155 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
3156 else if (clrprealloc)
3157 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
3159 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3160 xfs_trans_set_sync(tp);
3162 error = xfs_trans_commit(tp, 0);
3164 xfs_iunlock(ip, XFS_ILOCK_EXCL);